v1.1.15

[1.1.15] — 2026-07-05

OpenAI-compatible provider fidelity release (part three) and Cloudflare. The sixth provider-readiness remediation phase aligns the AI21, Cloudflare, DeepInfra, Qwen, and SambaNova providers on request/response correctness. No breaking API changes relative to v1.1.14.

Fixed

• AI21 Jamba errors and usage: the Jamba chat and streaming paths now route through the shared OpenAI-compatible path, so errors, finish reasons, and streaming token usage are handled consistently.
• Error message extraction: a gateway-level error returned as a {"detail":"…"} body (e.g. AI21 authentication errors) now surfaces the extracted message instead of the raw JSON body.
• AI21 model list: refreshed to the current Jamba 1.7 generation; the deprecated Jurassic /complete path is retained but marked legacy.
• Cloudflare embeddings: now route through the shared embeddings path, so errors surface consistently and input is normalized.

Changed

• Live model discovery added for DeepInfra and SambaNova (self-updating model lists).
• SambaNova embeddings: SambaNova now supports the embeddings endpoint via the shared helper.
• Base-URL validation is now enforced at construction for AI21, Cloudflare, DeepInfra, Qwen, and SambaNova.

Contributors

Thanks to everyone who shipped this release:

• v1.1.15 - OpenAI-Compat Providers III & Cloudflare (#326) — @MitulShah1

---

Full changelog: https://github.com/ferro-labs/ai-gateway/blob/v1.1.15/CHANGELOG.md

Thanks to @viniciusmtsantos for the qwen error-path test coverage (#325).